Yang Shouxin, Wu Ning, Zhang Li, Li Meng
Department of Diagnostic Radiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
PET-CT Center, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
Ann Transl Med. 2021 Feb;9(4):279. doi: 10.21037/atm-20-2992.
To investigate the influence of slice thickness on radiomic feature (RF) values in solid pulmonary nodules and evaluate the effect of a linear interpolation method in correcting the influence.
Thirty pulmonary nodules from 28 patients were selected prospectively with a thick-slice of 5 mm and a thin-slice of 1.25 mm on CT. A resampling method was used to normalize the voxel size of thick and thin slices CT images to 1×1×1 mm by linear interpolation. Lung nodules were segmented manually. A total of 396 radiomic features (RFs) were extracted from thick-slice and thin-slice images, together with the images resampled from thick (thick-r) and thin (thin-r) slices. The differences between the RF values were evaluated using a paired -test. A comparison between groups was made using the Chi-squared test.
Among the 396 RFs, 305 RFs showed an intraclass correlation coefficient ≥0.75 after test-retest analysis (including 22 histogram features, 20 geometry features, and 263 texture features). In the non-resampled data, 239 RF values (78.4%, 239/305) showed significant differences between thick and thin slice CT images. Resampling of thick images revealed that 202 RF values (66.2%, 202/305) showed significant differences between thick-r and thin slice CT images, showing a significant decrease in the number of different RF values when compared to non-resampled data (P<0.01). For the RF subgroups, only texture features showed a significant reduction in the number of different RF values after resampling (P<0.01). When both thick and thin slice images were resampled, the number of significantly different RF values between thick-r and thin-r images was increased to 247 (81.0%, 247/305), showing no significant difference when compared to non-resampled data (P=0.421).
Slice thickness demonstrated a considerable influence on RF values in solid pulmonary nodules, producing false results when CT images with different slice thicknesses were used. Linear interpolation of the resampling method was limited because of the relatively small correction effect.
探讨层厚对实性肺结节中放射组学特征（RF）值的影响，并评估线性插值法在纠正该影响方面的效果。
前瞻性选取28例患者的30个肺结节，CT扫描时分别采用5mm厚层和1.25mm薄层扫描。采用重采样方法通过线性插值将厚层和薄层CT图像的体素大小归一化为1×1×1mm。手动分割肺结节。从厚层图像、薄层图像以及厚层重采样（厚-r）和薄层重采样（薄-r）图像中提取总共396个放射组学特征（RFs）。使用配对t检验评估RF值之间的差异。采用卡方检验进行组间比较。
在396个RFs中，经过重测分析后，305个RFs的组内相关系数≥0.75（包括22个直方图特征、20个几何特征和263个纹理特征）。在未重采样的数据中，239个RF值（78.4%，239/305）在厚层和薄层CT图像之间显示出显著差异。厚层图像重采样后显示，202个RF值（66.2%，202/305）在厚-r和薄层CT图像之间显示出显著差异，与未重采样数据相比，不同RF值的数量显著减少（P<0.01）。对于RF亚组，只有纹理特征在重采样后不同RF值的数量显著减少（P<0.01）。当厚层和薄层图像都进行重采样时，厚-r和薄-r图像之间显著不同的RF值数量增加到247个（81.0%，247/305），与未重采样数据相比无显著差异（P=0.421）。
层厚对实性肺结节的RF值有相当大的影响，使用不同层厚的CT图像会产生错误结果。重采样方法的线性插值由于校正效果相对较小而受到限制。
